Fork: 1084 Star: 8810 (更新于 2024-07-19 13:19:39)

license: Apache-2.0

Language: Python .

Train transformer language models with reinforcement learning.

最后发布版本: v0.9.6 ( 2024-07-08 21:51:10)

官方网址 GitHub网址

TRL - Transformer Reinforcement Learning

Full stack library to fine-tune and align large language models.

License Documentation GitHub release

What is it?

The trl library is a full stack tool to fine-tune and align transformer language and diffusion models using methods such as Supervised Fine-tuning step (SFT), Reward Modeling (RM) and the Proximal Policy Optimization (PPO) as well as Direct Preference Optimization (DPO).

The library is built on top of the transformers library and thus allows to use any model architecture available there.


  • Efficient and scalable:
    • accelerate is the backbone of trl which allows to scale model training from a single GPU to a large scale multi-node cluster with methods such as DDP and DeepSpeed.
    • PEFT is fully integrated and allows to train even the largest models on modest hardware with quantisation and methods such as LoRA or QLoRA.
    • unsloth is also integrated and allows to significantly speed up training with dedicated kernels.
  • CLI: With the CLI you can fine-tune and chat with LLMs without writing any code using a single command and a flexible config system.
  • Trainers: The Trainer classes are an abstraction to apply many fine-tuning methods with ease such as the SFTTrainer, DPOTrainer, RewardTrainer, PPOTrainer, CPOTrainer, and ORPOTrainer.
  • AutoModels: The AutoModelForCausalLMWithValueHead & AutoModelForSeq2SeqLMWithValueHead classes add an additional value head to the model which allows to train them with RL algorithms such as PPO.
  • Examples: Train GPT2 to generate positive movie reviews with a BERT sentiment classifier, full RLHF using adapters only, train GPT-j to be less toxic, StackLlama example, etc. following the examples.


Python package

Install the library with pip:

pip install trl

From source

If you want to use the latest features before an official release you can install from source:

pip install git+


If you want to use the examples you can clone the repository with the following command:

git clone

Command Line Interface (CLI)

You can use TRL Command Line Interface (CLI) to quickly get started with Supervised Fine-tuning (SFT), Direct Preference Optimization (DPO) and test your aligned model with the chat CLI:


trl sft --model_name_or_path facebook/opt-125m --dataset_name imdb --output_dir opt-sft-imdb


trl dpo --model_name_or_path facebook/opt-125m --dataset_name trl-internal-testing/hh-rlhf-helpful-base-trl-style --output_dir opt-sft-hh-rlhf 


trl chat --model_name_or_path Qwen/Qwen1.5-0.5B-Chat

Read more about CLI in the relevant documentation section or use --help for more details.

How to use

For more flexibility and control over the training, you can use the dedicated trainer classes to fine-tune the model in Python.


This is a basic example of how to use the SFTTrainer from the library. The SFTTrainer is a light wrapper around the transformers Trainer to easily fine-tune language models or adapters on a custom dataset.

# imports
from datasets import load_dataset
from trl import SFTTrainer

# get dataset
dataset = load_dataset("imdb", split="train")

# get trainer
trainer = SFTTrainer(

# train


This is a basic example of how to use the RewardTrainer from the library. The RewardTrainer is a wrapper around the transformers Trainer to easily fine-tune reward models or adapters on a custom preference dataset.

# imports
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from trl import RewardTrainer

# load model and dataset - dataset needs to be in a specific format
model = AutoModelForSequenceClassification.from_pretrained("gpt2", num_labels=1)
tokenizer = AutoTokenizer.from_pretrained("gpt2")


# load trainer
trainer = RewardTrainer(

# train


This is a basic example of how to use the PPOTrainer from the library. Based on a query the language model creates a response which is then evaluated. The evaluation could be a human in the loop or another model's output.

# imports
import torch
from transformers import AutoTokenizer
from trl import PPOTrainer, PPOConfig, AutoModelForCausalLMWithValueHead, create_reference_model
from trl.core import respond_to_batch

# get models
model = AutoModelForCausalLMWithValueHead.from_pretrained('gpt2')
ref_model = create_reference_model(model)

tokenizer = AutoTokenizer.from_pretrained('gpt2')
tokenizer.pad_token = tokenizer.eos_token

# initialize trainer
ppo_config = PPOConfig(batch_size=1, mini_batch_size=1)

# encode a query
query_txt = "This morning I went to the "
query_tensor = tokenizer.encode(query_txt, return_tensors="pt")

# get model response
response_tensor  = respond_to_batch(model, query_tensor)

# create a ppo trainer
ppo_trainer = PPOTrainer(ppo_config, model, ref_model, tokenizer)

# define a reward for response
# (this could be any reward such as human feedback or output from another model)
reward = [torch.tensor(1.0)]

# train model for one step with ppo
train_stats = ppo_trainer.step([query_tensor[0]], [response_tensor[0]], reward)


DPOTrainer is a trainer that uses Direct Preference Optimization algorithm. This is a basic example of how to use the DPOTrainer from the library. The DPOTrainer is a wrapper around the transformers Trainer to easily fine-tune reward models or adapters on a custom preference dataset.

# imports
from transformers import AutoModelForCausalLM, AutoTokenizer
from trl import DPOTrainer

# load model and dataset - dataset needs to be in a specific format
model = AutoModelForCausalLM.from_pretrained("gpt2")
tokenizer = AutoTokenizer.from_pretrained("gpt2")


# load trainer
trainer = DPOTrainer(

# train


If you want to contribute to trl or customizing it to your needs make sure to read the contribution guide and make sure you make a dev install:

git clone
cd trl/
make dev


Proximal Policy Optimisation

The PPO implementation largely follows the structure introduced in the paper "Fine-Tuning Language Models from Human Preferences" by D. Ziegler et al. [paper, code].

Direct Preference Optimization

DPO is based on the original implementation of "Direct Preference Optimization: Your Language Model is Secretly a Reward Model" by E. Mitchell et al. [paper, code]


  author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang},
  title = {TRL: Transformer Reinforcement Learning},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{}}

最近版本更新:(数据更新于 2024-07-19 13:19:23)

2024-07-08 21:51:10 v0.9.6

2024-06-06 22:17:27 v0.9.4

2024-06-06 00:08:05 v0.9.3

2024-04-22 16:59:58 v0.8.6

2024-04-18 19:58:41 v0.8.5

2024-04-17 23:22:10 v0.8.4

2024-04-12 18:25:23 v0.8.3

2024-04-11 21:51:28 v0.8.2

2024-03-20 18:39:16 v0.8.1

2024-03-20 00:25:11 v0.8.0

huggingface/trl同语言 Python最近更新仓库

2024-07-21 02:15:32 ultralytics/ultralytics

2024-07-20 08:04:16 marimo-team/marimo

2024-07-20 06:03:39 mem0ai/mem0

2024-07-20 01:38:59 home-assistant/core

2024-07-19 19:26:44 xorbitsai/inference

2024-07-19 16:01:49 zhayujie/chatgpt-on-wechat